Diagnosis of Parkinson Disease Using Ensemble Methods for Class Imbalance Problem

Parkinson disease (PD) is the most prevalent degenerative neurological disorders that is incurable. Early PD diagnosis is essential in order to determine the initial course of treatment. Typically, the issue of class imbalance has an impact on the PD diagnosis. This paper seeks to give a comparative analysis of the ensemble methods: random forest, bagging, and random under-sampling boost for addressing the class imbalance problem for PD diagnosis. We make use of a real-world PD speech dataset that is housed in the repository at UCI (University of California, Irvine). Due to the high imbalance in this dataset, feature scaling and the Synthetic Minority Oversampling Technique (SMOTE) are employed. We also employ the feature selection (FS) technique for enhancing the efficiency of the machine learning algorithms (MLAs). The results show that bagging performs best with an accuracy of 96.46%. This study proposes the use of ensemble approaches for PD’s early diagnosis.